Assessing the Impacts of Livestock Production on Biodiversity in Rangeland Ecosystems

Assessing the impacts of livestock production on biodiversity in rangeland ecosystems

Rob Alkemadea,1, Robin S. Reidb, Maurits van den Berga, Jan de Leeuwc, and Michel Jeukena

aPBL Netherlands Environmental Assessment Agency, 3720 AH Bilthoven, The Netherlands; bCenter for Collaborative Conservation, Colorado State University, Fort Collins, CO 80523; and cInternational Livestock Research Institute, Nairobi 00100, Kenya

Edited by Mario Herrero, International Livestock Research Institute, Nairobi, Kenya, and accepted by the Editorial Board August 30, 2011 (received for review December 15, 2010) Biodiversity in rangelands is decreasing, due to intense utilization reindeer, and other ungulates (10). The effects of grazing on for livestock production and conversion of rangeland into crop- rangeland biodiversity include the removal of biomass, trampling land; yet the outlook of rangeland biodiversity has not been and destruction of root systems, and replacement of wild grazers considered in view of future global demand for food. Here we by livestock (6). The combined effects depend on (i) the extent assess the impact of future livestock production on the global of rangelands grazed by livestock, (ii) the grazing intensity, (iii) rangelands area and their biodiversity. First we formalized exist- the original type of vegetation (e.g., impacts are greater when ing knowledge about livestock grazing impacts on biodiversity, forests are cleared for the purpose of grazing), and (iv) land expressed in mean species abundance (MSA) of the original management (e.g., fertilized, planted with exotic species, etc.). rangeland native species assemblages, through metaanalysis of The extent of rangelands has changed over time due to con- peer-reviewed literature. MSA values, ranging from 1 in natural version of forested land into man-made grasslands (e.g., ref. 11), rangelands to 0.3 in man-made grasslands, were entered in the the conversion of rangeland into cropland, and the replacement IMAGE-GLOBIO model. This model was used to assess the impact of abandoned rangeland with forests (e.g., ref. 12). As a result of of change in food demand and livestock production on future these processes, the extent of grazed rangelands increased by rangeland biodiversity. The model revealed remarkable regional 24,000 km2 annually, between 1987 and 2000 (13). The rates of variation in impact on rangeland area and MSA between two conversion of land and the intensity of rangeland use are likely to agricultural production scenarios. The area of used rangelands continue changing over the next decades as a result of the slightly increases globally between 2000 and 2050 in the baseline forecasted increased global demand for food. However, the im- scenario and reduces under a scenario of enhanced uptake of pact on rangeland biodiversity of such increases in demand for resource-efficient production technologies increasing production crop and livestock commodities through possible conversion to [high levels of agricultural knowledge, science, and technology and from rangeland and changes in the intensity of their use for (high-AKST)], particularly in Africa. Both scenarios suggest a global livestock production remain poorly understood. decrease in MSA for rangelands until 2050. The contribution of Over the last century, increased livestock production has been livestock grazing to MSA loss is, however, expected to diminish achieved mainly through a shift from pastoral systems with free- after 2030, in particular in Africa under the high-AKST scenario. range feeding toward mixed and industrial systems, where Policies fostering agricultural intensification can reduce the overall a substantial part (>10%) of the feed comes from crops or crop pressure on rangeland biodiversity, but additional measures, by-products and so-called landless or industrial livestock pro- addressing factors such as climate change and infrastructural de- duction systems, where the bulk of the feed (>90%) is produced velopment, are necessary to totally halt biodiversity loss. off farm. Consequences of these shifts are substantial increases in cropland area for feed production and a strong increase in dose-response model | intactness | land use animal population densities outside rangelands (14). Although this change may have released pressure on rangeland systems, it abitat degradation and land use change are among the has not avoided expansion of domestic livestock grazing into Hmajor factors causing biodiversity loss, worldwide (1). natural rangelands. This outcome has been the case in most of Rangelands are no exception as they are under stress from Africa and in the Brazilian Cerrado and Amazon region, whereas conversion into cropland and pressure from livestock and ex- at the same time grazed rangelands in the central south of Brazil cessive fire (2, 3). Other important drivers of change include were replaced by cropland, such as for soybean and sugarcane climate change, habitat fragmentation, and the development of production (15, 16). infrastructure (4–6). Rangelands, which compose ∼25% of the These trends are expected to continue in the future, as the ∼ world’s land area, include grasslands, scrublands, woodlands, human population will increase to 9.3 billion by 2050 (17). This wetlands, and deserts and are found from the Asian steppes to increase, combined with increasing prosperity and a shift in di- the Andean regions of South America and from the mountains etary patterns, will lead toward a larger demand for meat and milk. This trend may be enhanced when costs of animal products of Western Europe to the African savanna. Vegetation and wild herbivores coevolved in many rangeland fall if production shifts from traditional pastoral systems to more energy- and nutrient-efficient production systems (17). systems and some are among the most species-rich in the world. Several approaches are available to quantify impacts of land The scrublands of the Southern African Karoo, for example, use on biodiversity (1). A frequently used method is to project support >6,000 different plant species (7), whereas African savannas are reputed for their rich diversity of large mammals. However, losses of wildlife in African rangelands are increasingly Author contributions: R.A. designed research; R.A. and M.J. performed research; R.A., attributed to encroachment of agriculture and competition with M.v.d.B., and M.J. analyzed data; and R.A., R.S.R., M.v.d.B., and J.d.L. wrote the paper. livestock (8, 9). The authors declare no conflict of interest. Livestock farming is the most widespread human activity and This article is a PNAS Direct Submission. M.H. is a guest editor invited by the Editorial the dominant land use in rangeland ecosystems. Rangelands Board. provide 10% of the global meat supply and support an estimated 1To whom correspondence should be addressed. E-mail: [email protected]. 200 million pastoralists and the herds of nearly 1 billion camel- This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. ids, cattle, and smaller livestock, in addition to yaks, horses, 1073/pnas.1011013108/-/DCSupplemental.

20900–20905 | PNAS | December 24, 2013 | vol. 110 | no. 52 www.pnas.org/cgi/doi/10.1073/pnas.1011013108 Downloaded by guest on September 30, 2021 future species distribution using empirical models describing the relation between environmental variables and individual species SPECIAL FEATURE (e.g., ref. 18). This approach assumes that the range of species will change with these environmental variables. Other authors use species–area relationships, assuming that the number of species directly relates to the extent of natural area (2, 19). The GLOBIO3 model uses general cause–effect relationships between human impact factors and a generic measure of intactness of biodiversity: the relative mean species abundance (MSA) of originally occurring species in pristine ecosystems (4). It ex- presses the relative difference between an undisturbed ecosystem and a disturbed or managed ecosystem on a scale from 1 (undisturbed) to 0 (completely destroyed) (20). GLOBIO3 allows the calculation of the effect of grazing within the context of multiple drivers of biodiversity change. The model currently accounts for the following impact factors: land use (including grazing), infrastructural development, habitat fragmentation, nitrogen deposition, and climate change. Fig. 1. Extent of used rangelands and man-made grasslands (open bars), Using the GLOBIO3 approach, this paper presents a first at- natural rangelands (shaded bars), and other vegetation (solid bars) for tempt to quantify biodiversity impacts of livestock production in biomes, derived from GLC2000 and IMAGE. rangeland ecosystems. First, the extent of rangeland ecosystems, including man-made grazing areas, was derived from the GLC2000 land-cover and land-use map and the Integrated The systematic literature review, which included 24 studies that Model to Assess the Global Environment (IMAGE) (21, 22). compared species composition in grazed systems with that in IMAGE provides estimates of the extent of rangelands used for natural rangelands, covered the full spectrum of rangeland livestock grazing and man-made pastures, here together called ecosystems (Table S1.1). An overall degree of land-use intensity grazed rangelands. Subsequently, we estimated a possible future was derived from these studies by using the reported grazing extent of land cover on the basis of the projection by the In- intensity, e.g., based on relative stocking rates, rangeland man- ternational Assessment of Agricultural Knowledge, Science, and agement, visual alteration of the vegetation structure, and sea- Technology for Development (IAASTD) (17), for the 2000–2050 sonal variation in grazing (Table S1.2). We assumed that authors period. In addition, we analyzed a variant of this projection, in report grazing intensity on the basis of their knowledge of the which a high scientifically based level of productivity increase is systems studied and that their classifications of grazing intensities assumed: the high levels of agricultural knowledge, science, and implicitly account for carrying capacity. The overview of all 24 technology (high-AKST) variant (23). Then we derived MSA studies resulted in five different categories, including “ungrazed,

values for a gradient of grazing intensities on rangelands by abandoned rangeland” where domesticated grazers were ex- SCIENCES

conducting a systematic literature review and metaanalysis (24). cluded, wildlife grazers did not recover, and forests did not de- ENVIRONMENTAL These MSA values were included in the GLOBIO3 model (4), velop; “natural rangeland” with wildlife or livestock grazing at enabling the analysis of changing grazing pressure on bio- stocking rates close to natural; “moderately used rangelands”; diversity. We used the GLOBIO3 model to estimate current and “intensively used rangelands”;and“man-made grasslands” future total MSA values, in relation to future developments (Table 1). according to the IAASTD baseline and the high-AKST scenario We used this metaanalysis for estimating MSA values for each variant. In addition, we explored the sensitivity of the outcomes of the grazing intensity categories (Table S2.1). The MSA value to the variability of MSA values. for natural rangelands was set at 1 (Methods). The MSA values SCIENCE for the other categories were statistically tested against the hy- SUSTAINABILITY Results pothesis of having no effect. All categories showed a value sta- Current rangelands occur in all biomes, but primarily in the grasslands and steppe, savanna, scrubland, and tundra. Signiﬁ- cant parts of tropical forest and temperate forest biomes also consist of, mostly man-made, rangeland (Fig. 1). In general, between 10% and 60% of the existing rangeland is used for livestock grazing (Fig. 1). The IAASTD baseline projection indicates an increase in total grazed rangeland, i.e., used and man-made rangelands, area (Fig. 2) of just over 1 million km2 (+5%) between 2000 and 2030, followed by a decrease of 0.8 million km2 between 2030 and 2050, resulting in a marginal 1% net increase between 2000 and 2050. The area of natural rangelands will decline from 25 million km2 to 22 million km2 (−11%) from 2000 to 2050. Together, these trends reﬂect the hypothesis that livestock production will become less dependent on free-roaming grazing and increasingly reliant on feed produced by cropland systems. The high-AKST scenario, with agricultural productivity increasing beyond the baseline assumptions, results in a decrease of the extent of grazed rangeland and an increase in the area of Fig. 2. Projections of future areas of natural rangelands for the IAASTD natural rangelands (Fig. 2). This effect is particularly strong in baseline scenario (solid line) and the high-AKST scenario (dotted line) and of Africa, where, according to this scenario, the area of land grazed grazed rangelands (including man-made grasslands) for the IAASTD baseline by livestock will have decreased by almost 50% by 2050, thus scenario (dashed lines) and the high-AKST scenario (dashed-dotted line), increasing the extent of natural rangelands (Figs. 3 and 4). derived from ref. 28.

Alkemade et al. PNAS | December 24, 2013 | vol. 110 | no. 52 | 20901 Downloaded by guest on September 30, 2021 impacts on MSA in the Americas, whereas other regions present a slight gain in the livestock-related MSA component (Fig. 6) as a consequence of the contraction of grazed rangelands (i.e., parts of the more intensively used rangelands are converted to cropland, resulting in an average MSA increase of the remaining rangelands). The high-AKST variant suggests little or no change in the effects of livestock grazing on rangeland MSA between 2000 and 2050 for most regions, except for Africa, where MSA will be significantly higher than in the baseline scenario (Fig. 6). However, these regional trends will not stop the overall global decline in MSA, because of other factors that are expected to persist, indicated by the decreasing total MSA values shown in Fig. 5. The MSA decrease in Europe in the high-AKST variant, both compared with the baseline as well as with the 2000 situation (Fig. 6), is mainly due to the abandonment of some of the most extensively used European grassland areas, which, however, would then evolve in forest rather than rangeland. Average grazing pressure in the remaining, more intensively used, ran- Fig. 3. Extent of grazed rangeland (including man-made grasslands) in each world region according to the baseline scenario in 1970 (solid bars), 2000 gelands would thus be higher compared with the baseline or with (bars with dark shading), 2030 (bars with light shading), and 2050 (open bars). the 2000 situation. Discussion tistically significant below 1 (P < 0.05) (Table 1 and Table S2.1). The area of rangeland under livestock grazing has expanded in MSA values decreased with increasing intensity, from 0.6 in the past and, according to the IAASTD reference scenario, will moderately used rangelands to 0.3 in man-made grasslands. continue to increase, albeit at a moderate pace, in some regions Ungrazed, abandoned rangelands presented an average MSA of of the world, especially in Central and South America. In the rest 0.7. No significant regional differences were found for MSA of the world, the area of land under grazing management is values and no differences between taxa were found (Table S2.2). expected to remain stable or even decrease. Under the same These values were subsequently entered in the GLOBIO3 model scenario livestock productivity increases, e.g., ruminant meat and to calculate changes in MSA values in the projections of the milk production increases by 118% and 66%, respectively, over future (4). the same period. Increased livestock production will depend The GLOBIO model suggests that biodiversity, in terms of more and more on feed produced on croplands. Cropland areas, global average MSA, will continue to decline, between 2000 and therefore, are expected to expand (14), although the bulk of the 2050, in rangeland ecosystems in both the baseline scenario and, increased crop production and livestock products is assumed to to a lesser extent, in the high-AKST variant (Fig. 5). MSA loss in come from an increase in agricultural productivity, as has been rangeland ecosystems is caused by a combination of pressure the case in the past. This result means that the environmental factors, including climate change, cropland expansion, frag- impacts of livestock will increasingly be associated with cropland fi mentation, infrastructural development, and livestock grazing. expansion and crop production intensi cation. The area of nat- At the global level, the impact of livestock grazing (i.e., ignoring ural rangelands is expected to decrease up to 2030 and sub- all other pressure factors) is expected to increase until 2030 and sequently remain almost stable under baseline conditions. Substantial areas could potentially be restored under conditions then decrease again, most notably in the high-AKST variant. fi Due to variability of the MSA estimates this difference at the favoring the development and implementation of resource-ef - global level is only marginal. In the baseline scenario, the ex- cient production technologies, as illustrated for the high-AKST pansion of livestock grazing causes an increase of grazing variant. This variant assumes a ruminant meat and milk productivity increase of >30% relative to that in the baseline scenario. These projections reflect an accelerated replacement of pastoral livestock production by production in more resource- efficient mixed systems. Note that the model does take account of price effects, which result in higher consumption (and thus, production) levels in the high-AKST scenario than in the baseline scenario. The high-AKST scenario does not imply a com- plete wipeout of traditional livestock production systems. First, the trend described above does not include the areas of very extensively used seminatural grasslands, as, in both scenarios, these grasslands remain relatively untouched. Second, even when these areas are excluded, land-use intensity in Africa, in terms of output of meat or milk per hectare of managed rangeland, would still be <40% of the average of that of the EU27 countries in the year 2000. The area of arable land in the high-AKST scenario is larger than that in the year 2000, but considerably smaller than in the baseline scenario. The overall balance of the opposing effects of the high-AKST variant on rangelands in different regions demonstrates the complexity of the processes involved and the importance of fol- Fig. 4. Extent of grazed rangeland (including man-made grasslands) in each lowing an integrated approach. world region according to the high-AKST scenario in 1970 (solid bars), 2000 It must be appreciated, though, that also in regions where the (bars with light shading), 2030 (bars with dark shading), and 2050 (open bars). model results suggest clear environmental benefits of high-AKST

20902 | www.pnas.org/cgi/doi/10.1073/pnas.1011013108 Alkemade et al. Downloaded by guest on September 30, 2021 Table 1. Average MSA values for different grazing intensities, resulting from the metaanalysis SPECIAL FEATURE Short description MSA SE

Natural rangelands Rangeland ecosystems determined by climatic and geographical circumstances and grazed by 1.0 — wildlife or domestic animals at rates similar to those of free-roaming wildlife Moderately used Rangelands with higher stocking rates: grazing has different seasonal patterns or vegetation 0.6 0.04 rangelands structure is different compared with natural rangelands Intensively used Rangelands with very high stocking rates: grazing has different seasonal patterns and 0.5 0.06 rangelands vegetation structure is different compared with natural rangelands Man-made grasslands Rangeland with high degree of human management, including converted forests 0.3 0.08 Ungrazed abandoned Original grasslands no longer in use, lacking wildlife grazing and no forests developed 0.7 0.07 rangelands

developments, such developments also have some inherent risks development and implementation of high-yielding agricultural and challenges that need to be addressed: practices are an important component, but are not sufficient for tackling rangeland biodiversity loss. The reverse is also true. Although yield increases such as those assumed for the high- Policies aimed at the preservation of rangeland biodiversity are AKST scenario are biophysically and technically possible (25, deemed to fail if they do not concurrently address the challenge 26), their realization is a major challenge, especially in regions of meeting future food demands. where they could be most beneficial, such as in sub-Saharan Nature in rangeland ecosystems differs considerably between Africa, but where progress is hindered by a combination of world regions. In northern and central Europe most grasslands socioeconomic and institutional constraints that need to be used for grazing are man made, developed from former forests. tackled concurrently (10, 27, 28). These ancient systems are considered seminatural grasslands and Although yield increases are a precondition for preserving valued as one of the most species-rich ecosystems in this region land for nature without compromising food security, these (12, 34). Large areas of seminatural rangelands either have been increases are not automatically achieved (29). converted into croplands during the past century or have been Restoration of nature in abandoned areas by natural succes- abandoned. Conservation and restoration of seminatural grass- sion may lead to land degradation if no specific interventions lands is one of the main objectives of current biodiversity policies are set in place (30). throughout Europe. In many world regions, forests are being Although increased yields, overall, result in improved food converted into either cropland or man-made grasslands, espe- security, some groups of farmers and farm workers are likely cially in Latin America, where livestock grazing is a primary to become negatively affected. Traditional pastoral culture cause of deforestation (e.g., refs. 11 and 35). In many other and the rich knowledge that pastoral people have about ways regions, livestock grazing has developed on natural rangeland, to adapt to climate change, for example, are in danger of such as grass-dominated systems or open woodlands (6). These SCIENCES being lost. grazed rangelands range from traditional pastoral systems to ENVIRONMENTAL Intensification based on high AKST is not a matter of doing farm settlements and fenced-in grasslands. more of the same on a smaller area of land. It must be an eco- We show that shifts in livestock production would have a ma- efficient intensification, i.e., covering the interrelationships jor impact on biodiversity change in rangeland ecosystems. and trade-offs among production, conservation, economic, Grazing intensity is difficult to measure directly as it depends on and social values (31–33). the relative stocking rate of livestock, the carrying capacity of the

ecosystem, and additional management of the land. We derived SCIENCE

a grazing intensity gradient from studies that were selected for SUSTAINABILITY Although in-depth discussion of these challenges is beyond the metaanalysis. Many authors provide only a qualitative de- scope of this paper, it should be clear that policies targeting the scription, as they focus on comparison between sites (Table S1.2 and literature cited therein). The lack of quantitative indicators for grazing intensities makes the assessment of the impact of grazing difﬁcult. The results of our study partly depend on the accuracy of MSA estimates derived from the metaanalysis. Due to the limitation of data availability, the estimated MSA values are not region spe- ciﬁc and bear considerable uncertainty. The sensitivity analysis shows that the results are robust enough to draw the following general conclusions. The projected changes in biodiversity, in terms of MSA, show a continual decreasing trend over the coming 40 y, in both scenarios, whereas the impact of grazing is expected to be reduced (Fig. 5). This result is mainly because the reduced impact by grazing is offset by increasing impacts by factors such as climate change, nitrogen deposition, and infrastructural development. The effect of livestock grazing is expected to decrease, especially in the high-AKST variant. In Africa, where the area of used rangelands decreases considerably under high AKST, natural Fig. 5. Global average MSA estimates, considering all environmental factors for the baseline scenario (solid line) and the high-AKST scenario (dotted rangelands can be restored, and MSA values show substantial line) and the effect on MSA of livestock grazing only, for the baseline improvement compared with the baseline scenario (Fig. 6). In (dashed lines) and the high-AKST scenario (dashed-dotted line). The error Latin America, where the conversion of nature to man-made bars indicate 2 × SD derived from the variability of MSA values. grasslands is avoided in the high-AKST variant, no further de-

Alkemade et al. PNAS | December 24, 2013 | vol. 110 | no. 52 | 20903 Downloaded by guest on September 30, 2021 or abandoned (0) natural grazing (1), moderate grazing intensity (2), high grazing intensity (3); visual alteration of the vegetation structure, not or slightly altered (0), signiﬁcantly altered in height or species composition, including exotics (1); rangeland management, no management (0), presence of management such as soil disturbance, clearance of vegetation and ap- plication of fertilizers, planting or sowing grass or feed crops (1); and seasonal variation, only seasonal grazing corresponding to natural grazing pattern (0), continuous grazing regardless of the season (1). The four indicators were combined into four grazing intensity classes, using the following rules: If the reported intensity of rangeland management equals 0, and the description is clear on the absence of wildlife grazing, e.g., by fenced enclosures, then the land is regarded as “ungrazed”, abandoned rangeland (36, 37); if the reported intensity equals 1, then the intensity class is that of man-made grasslands (38, 39); if the sum of the reported intensity, visual alteration of the vegetation structure, and seasonal variation equals 1, then grazing is regarded as “natural” (40–42); if this sum is 2 or 3, then the land is considered moderately used grazing land (43, 44); and if the sum is 4 or 5, then the intensity class is intensive (44, 45). Each dataset consisted of samples from locations with different grazing Fig. 6. Changes in the effects of grazing pressure on rangeland MSA from intensities, containing observations from different points or subsamples. The 2000 to 2050, according to the baseline scenario (bars with dark shading) observations were summarized to a mean or sum of individuals found for and the high-AKST variant (bars with light shading). Error bars indicate ±1 each species. For each study k, each species i, and each grazing intensity e the

SD, derived from variability in MSA values. abundance is denoted as nike with variance V(nike). The idea behind MSA is to describe the relative difference between the original ecosystem and the disturbed state of the same ecosystem. So only the species found in samples cline of MSA values is expected. The situation will remain ap- from the natural rangeland were included. proximately stable in North America and North Asia and may Ratios of the abundance of single species found in disturbed and un- have contradictory effects in Europe, where high-AKST values disturbed ecosystems are the basis on which to calculate MSA. Disturbances of decrease compared with the baseline (Fig. 6). ecosystems often lead to the decline of many species and simultaneous This result suggests that policies that foster high agricultural increases in others. Both decreases and increases signal disturbance. To avoid averaging out the relative changes of declining species by the increasing growth in croplands and a shift toward higher livestock pro- species, we capped the ratios at 1, so surpluses of individuals from the dis- ductivity in mixed crop livestock systems release the pressure on turbed ecosystem would be ignored. The ratio rike is calculated by dividing biodiversity in rangeland ecosystems in regions where pro- nike, the abundance of a species in a disturbed ecosystem by the abundance ductivity is still low. By these policies the conversion of range- found in the original ecosystem, niko if nike < niko; otherwise rike = 1. For each lands into cropland can also be slowed down. In particular, the study the ratios rike are summarized by taking the overall mean Rke.The African rangelands will be positively affected. In regions where MSAe for each grazing intensity e is calculated from all Rke derived from the technology is advanced and productivity is already high no pos- individual studies, P itive effect of increased productivity on biodiversity can be ð = Þ ¼ ¼ Pk Rke Vke ; expected. MSAe Re 1=Vke Despite the positive effects of increased productivity on bio- k diversity, other drivers, such as climate change, infrastructural where Vke is the pooled variance of the ratios of species abundances for each development, and conversion to croplands, continue to affect study and copes for differences between studies. For more detail see SI rangeland biodiversity. An integrated approach including poli- Methods, section 1. cies for multiple drivers may avoid the result that positive effects A random-effects metaanalysis was done, using the R 2.12.0 software, to of policies targeted at one driver are offset by negative effects of derive a pooled effect size for the different grazing intensities. Linear mixed- effect models were estimated for MSA to check for differences between another driver. regions and taxa. Methods Rangeland Extent and Intensity. The extent of rangeland ecosystems is diffi- The relation between grazing intensity and biodiversity was quantified using cult to estimate, as it depends on the definition of rangelands and the data from peer-reviewed literature selected through a systematic literature resolution of measurements. Global estimates depend on remotely sensed search. The effect of grazing is described in terms of relative mean species data and as the technology rapidly evolves, measurements of land cover type abundance (MSA) of originally occurring species, which compares the species also change. We estimated rangeland extent from the Global Land Cover composition of livestock grazing systems with adjacent natural systems, for database of 2000 (GLC2000) (21). The GLC2000 categories of rangeland example wildlife reserves (4). ecosystems are “shrub cover, closed–open, evergreen”; “shrub cover, closed– Studies were collected in two batches. The first batch contained literature open, deciduous”; “herbaceous cover, closed–open”; “sparse herbaceous or entirely based on a former review for the GLOBIO3 model (4), yielding 15 sparse shrub cover”; and “regularly flooded shrub and/or herbaceous papers on grazing intensity. The second batch was compiled from a search cover”. Extent of rangelands used for livestock grazing was estimated using using the keyword sequence species AND [grassland OR rangeland] AND livestock density maps from the Food and Agriculture Organization (Rome) graz* AND [intact OR natural OR primary OR virgin OR mature OR pristine] and assumptions on the area needed (46). In IMAGE, two intensity catego- AND [*diversity OR richness OR abundance*] of the SCOPUS database. This ries are considered: pastoralism and mixed grazing systems (25). These two search yielded 300 papers. We selected papers from 10 journals with >10 categories are assumed to correspond with the moderate grazing intensity papers that met the search criteria and rejected those that were obviously and high grazing intensity from this study. not of interest according to title and abstract. From the 49 remaining papers only those were selected that met the following criteria: Does the paper Scenarios. The baseline scenario in our study is the reference case of the contain data on species abundance? Could one of the land-use categories IAASTD (33). This baseline scenario was developed using several linked studied be considered natural or close to natural? Does the paper describe models, including, inter alia, the IMPACT agriculture-economy model (23) the effect of grazing? This procedure yielded an additional 9 papers con- and the IMAGE model (22). IMPACT is a partial equilibrium model, which taining sufficient data to calculate MSA values. Together with the 15 papers accounts for food demand, food production, resource availability, trade, and from the other batch, this procedure resulted in 24 papers, from which commodity prices. These results were then used as input in IMAGE (version datasets were derived and MSA values calculated (Table S1.1). Grazing in- 2.4) to compute the areas of land needed in various sectors and a number of tensity was extracted from the paper by scoring four indicators: reported other environmental parameters (22). The reference scenario depicts the intensity by author, sometimes with corresponding stocking rate, ungrazed world developing over the next decades in a business-as-usual mode, with-

20904 | www.pnas.org/cgi/doi/10.1073/pnas.1011013108 Alkemade et al. Downloaded by guest on September 30, 2021 out anticipating deliberate policy interventions. For population, the scenario The high-AKST variant assumes accelerated investment in the de-

is based on the United Nations medium projection, leading to a total pop- velopment and adoption of agricultural knowledge, science, and technol- SPECIAL FEATURE ulation of ∼9.2 billion by 2050, an almost 50% increase since 2000. Global ogy, leading to a worldwide extra 40% increase in crop yields (14) and an economic growth is assumed to be close to 3% annually, over the 2000–2050 extra 20% increase in livestock productivity, between 2005 and 2050, com- period, resulting in a doubling of GDP by 2050. Together, these drivers lead pared with in the baseline scenario. For livestock production it has been to an increasing food crop demand (an increase of ∼80% between 2000 and calculated that this result would lead to an extra dependence on feed 2050). Diets are projected to become more meat intensive, especially in low- from crops. income countries. Globally, meat demand increases by 115% between 2000 Biodiversity estimates at global and regional levels are estimated using and 2050, with annual growth rates of ∼1.7% (early in the scenario period) the IMAGE 2.4 model combined with the GLOBIO3 model (4). Both models to 1.4% (between 2025 and 2050). About 70% of crop production growth is are brieﬂy described in SI Methods, section 2. The impact of the variability of projected to come from yield increases, implying an expansion of cropland MSA values, derived from the metaanalysis, on the scenario outcomes was from 1.5 billion hectares, today, to >1.6 billion hectares, by 2050. The in- explored by varying the parameter setting in a Monte Carlo analysis. The crease in meat consumption is expected to lead to a signiﬁcant increase in variability of MSA values is due to, e.g., regional differences, taxonomical the number of animals. At the same time, however, there is a gradual shift differences, and differences in study design. SDs for global and regional from extensive to more intensive forms of animal husbandry. This shift MSA estimates were derived from this sensitivity analysis. We compared the implies that some net expansion of grazed rangelands and man-made outcomes per region between baseline and the high-AKST scenario with grasslands still occurs, but it levels off soon after 2025. Student’s t test, using unequal variances.

1. Pereira HM, et al. (2010) Scenarios for global biodiversity in the 21st century. Science 25. Koning NBJ, et al. (2008) Long-term global availability of food: Continued abundance 330:1496–1501. or new scarcity? NJAS Wageningen J Life Sci 55:229–292. 2. MEA (2005) Millennium Ecosystem assessment. Ecosystems and Human Well-being: 26. Lobell DB, Cassman KG, Field CB (2009) Crop yield gaps: Their importance, magni- Scenarios (Island Press, Washington, DC), Vol 2. tudes, and causes. Annu Rev Environ Res 34:179–204. 3. Leadley P, et al. (2010) Biodiversity scenarios: Projections of 21st century change in 27. InterAcademy Council (2004) Realizing the Promise and Potential of African Agri- biodiversity and associated ecosystem services. Technical Series no. 50, ed Secretariat culture: Science and Technology Strategies for Improving Agricultural Productivity of the Convention of Biological Diversity (Montreal). and Food Security in Africa (InterAcademy Council, Amsterdam). 4. Alkemade R, et al. (2009) GLOBIO3: A framework to investigate options for reducing 28. Izac AM, Egelyng H, Ferreira G (2009) Options for Enabling Policies and Regulatory global terrestrial biodiversity loss. Ecosystems (N Y) 12:374–390. Environments. Agriculture at a Crossroads—The Global Report. International Assess- 5. Dobson AP, et al. (2010) Road will ruin Serengeti. Nature 467:272–273. ment of Agricultural Knowledge, Science, and Technology, eds McIntyre BD, 6. Reid RS, et al. (2010) Global livestock impacts on biodiversity. Livestock in a Changing Herren HR, Wakhungu J, Watson RT (Island Press, Washington, DC), pp 441–494. Landscape. Drivers, Consequences, and Responses, eds Steinfeld H, Mooney HA, 29. Ewers RM, Scharlemann JPW, Balmford A, Green RE (2009) Do increases in agricultural – Schneider F, Neville LE (Island Press, Washington, DC), Vol 1, pp 111 138. yield spare land for nature? Glob Change Biol 15:1716–1726. 7. Conservation International (2007) Biodiversity Hotspots (Conservation International 30. Rey Benayas JM, Martins A, Nicolau JM, Schulz JJ (2007) Abandonment of agricultural Foundation), available at www.biodiversityhotspots.org. land: An overview of drivers and consequences. CAB Rev: Perspectives in Agriculture, 8. Msoffe FU, et al. (2011) Drivers and impacts of land use change in the Maasai Steppe Veterinary Science, Nutrition and Natural Resources 2:057. of Northern Tanzania: An ecological-social-political analysis. J Land Use Sci, 10.1080/ 31. Keating BA, et al. (2010) Eco-efﬁcient agriculture: Concepts, challenges, and oppor- 1747423x.2010.511682. tunities. Crop Sci 50:S-1–S-11. 9. Ogutu JO, Owen-Smith N, Piepho HP, Said MY (2011) Continuing wildlife population 32. Wilkins RJ (2008) Eco-efﬁcient approaches to land management: A case for increased – declines and range contraction in the Mara region of Kenya during 1977 2009. J Zool, integration of crop and animal production systems. Philos Trans R Soc Lond B Biol Sci 10.1111/j.1469-7998.2011.00818.x. 363:517–525. 10. Food and Agriculture Organization and World Bank (2009) Awakening Africa’s

33. Gurib-Fakim A, Smith L (2009) Options to enhance the impact of AKST on de- SCIENCES Sleeping Giant. Prospects for Commercial Agriculture in the Guinea Savannah Zone velopment and sustainability goals. Agriculture at a Crossroads—The Global Report.

and Beyond. (World Bank, Washington, DC; Food and Agriculture Organization, ENVIRONMENTAL International Assessment of Agricultural Knowledge, Science, and Technology, eds Rome). McIntyre BD, Herren HR, Wakhungu J, Watson RT (Island Press, Washington, DC), pp 11. Lambin EF, Geist HJ, Lepers E (2003) Dynamics of land-use and land-cover change in 377–440. tropical regions. Annu Rev Environ Res 28:205–241. 34. Pykälä J (2005) Plant species responses to cattle grazing in mesic semi-natural grass- 12. Laiolo P, Dondero F, Ciliento E, Rolando A (2004) Consequences of pastoral aban- land. Agric Ecosyst Environ 108:109–117. donment for the structure and diversity of the alpine avifauna. J Appl Ecol 41: 35. Nepstad DC, Stickler CM, Filho BS, Merry F (2008) Interactions among Amazon land 294–304. use, forests and climate: Prospects for a near-term forest tipping point. Philos Trans R 13. United Nations Environmental Programme (2007) Global Environmental Outlook 4: Soc Lond B Biol Sci 363:1737–1746.

Environment for Development (United Nations Environmental Programme, Nairobi, SCIENCE 36. Cagnolo L, Molina SI, Valladares GR (2002) Diversity and guild structure of insect Kenya). assemblages under grazing and exclusion regimes in a mountain grassland from SUSTAINABILITY 14. Bouwman AF, et al. (2011) Exploring global changes in nitrogen and phosphorus – cycles in agriculture induced by livestock production over the 1900–2050 period. Proc Central Argentina. Biodivers Conserv 11:407 420. 37. Hart RH (2001) Plant biodiversity on shortgrass steppe after 55 years of zero, light, Natl Acad Sci USA 110:20882–20887. – 15. Lapola DM, et al. (2010) Indirect land-use changes can overcome carbon savings from moderate, or heavy cattle grazing. Plant Ecol 155:111 118. biofuels in Brazil. Proc Natl Acad Sci USA 107:3388–3393. 38. Brotons L, Wolff A, Paulus G, Martin JL (2005) Effect of adjacent agricultural habitat – 16. Barona E, Ramankutty N, Hyman G, Coomes OT (2010) The role of pasture and soy- on the distribution of passerines in natural grasslands. Biol Conserv 124:407 414. bean in deforestation of the Brazilian Amazon. Environ Res Lett 5:024002. 39. Sánchez-Zapata JA, et al. (2003) Land use changes and raptor conservation in steppe – 17. McIntyre BD, Herren HR, Wakhungu J, Watson RT, eds (2009) International Assess- habitats of Eastern Kazakhstan. Biol Conserv 111:71 77. ment of Agricultural Knowledge, Science and Technology for Development. Global 40. Aumann T (2001) The structure of raptor assemblages in riparian environments in the – Report (IAASTD, Washington, DC). south-west of the Northern Territory, Australia. Emu 101:293 304. fi 18. Guisan A, Zimmermann NE (2000) Predictive habitat distribution models in ecology. 41. Bouyer J, et al. (2007) Identi cation of ecological indicators for monitoring ecosystem – Ecol Modell 135:147–186. health in the trans-boundary W Regional park: A pilot study. Biol Conserv 138:73 88. fi 19. Van Vuuren DP, Sala O, Pereira HM (2006) The future of vascular plant diversity under 42. Smart R, Whiting MJ, Twine W (2005) Lizards and landscapes: Integrating eld surveys four global scenarios. Ecol Soc 11(2):25. and interviews to assess the impact of human distrurbance on lizard assemblages and 20. Scholes RJ, Biggs R (2005) A biodiversity intactness index. Nature 434:45–49. selected reptiles in a savanna in South Africa. Biol Conserv 122:23–31. 21. Bartholome E, et al. (2004) Global Land Cover for the Year 2000 (European Com- 43. O’Connor TG (2005) Influence of land use on plant community composition and di- mission, Joint Research Centre, Ispra, Italy). versity in Highland Sourveld grassland in the southern Drakensberg, South Africa. J 22. Bouwman AF, Kram T, Klein Goldewijk K (2006) Integrated Modelling of Global En- Appl Ecol 42:975–988. vironmental Change. An Overview of IMAGE 2.4 [Netherlands Environmental As- 44. Woinarski JCZ, Ash AJ (2002) Responses of vertebrates to pastoralism, military land sessment Agency (MNP), Bilthoven, The Netherlands]. use and landscape position in an Australian tropical savanna. Austral Ecol 27:311–323. 23. Rosegrant M, et al. (2009) Final Report: Exploring Alternative Futures for Agricultural 45. Herremans M (1998) Conservation status of birds in Botswana in relation to land use. Knowledge, Science and Technology (AKST) (Australian Centre for International Biol Conserv 86:139–160. Agricultural Research, Canberra, Australia). 46. Klein Goldewijk K, Van Drecht G, Bouwman AF (2007) Mapping contemporary global 24. Pullin AS, Stewart GB (2006) Guidelines for systematic review in conservation and cropland and grassland distributions on a 5 × 5 minute resolution. J Land Use Sci 2: environmental management. Conserv Biol 20:1647–1656. 167–190.

Alkemade et al. PNAS | December 24, 2013 | vol. 110 | no. 52 | 20905 Downloaded by guest on September 30, 2021